This project aims to parse data from one of these two programs:
-
mosdepth --by ${intervals_bed} \ --thresholds 5,10,20,30 \ ${outut_prefix} ${bam}
-
java -jar /usr/gitc/GATK35.jar \ -T DiagnoseTargets \ -R ${ref_fasta} \ -o ${output_name} \ -I ${input_bam} \ -L ${intervals} \ -min ${minCov}
And organize it in a way that one could easily inspect how is the sequencing coverage in genes of interest.
Download and extract the proper release for your system, and then:
# For GATK DiagnoseTargets output
./coverage --vcf <coverage.vcf.gz> 1>stdout.tsv
# For Mosdepth '.thresholds.bed.gz' output
./coverage --bed <thresholds.bed.gz> 1>stdout.tsv
After that you'll have two files:
- stdout.tsv
ENSG00000187634 SAMD11 Count: 15 Total Bases: 3246 Coverage 5x: 2831 10x: 2514 20x: 2252 30x: 2137
ENSG00000188976 NOC2L Count: 19 Total Bases: 2250 Coverage 5x: 2250 10x: 2250 20x: 2250 30x: 2250
ENSG00000187961 KLHL17 Count: 12 Total Bases: 2134 Coverage 5x: 2071 10x: 2027 20x: 2012 30x: 1992
ENSG00000187583 PLEKHN1 Count: 15 Total Bases: 2327 Coverage 5x: 2327 10x: 2327 20x: 2316 30x: 2298
ENSG00000187642 PERM1 Count: 5 Total Bases: 2463 Coverage 5x: 2463 10x: 2463 20x: 2463 30x: 2447
...
- *_coverage.json.gz
"Identifier": "no_sample_id",
"GitVersion": "",
"GlobalTotalBases": 36726635,
"GlobalCoveredBases5x": 36462766,
"GlobalCoveredBases10x": 36290670,
"GlobalCoveredBases20x": 35708523,
"GlobalCoveredBases30x": 34512655,
"PerGeneCoverage": {
"ENSG00000000003": {
"ENSG": "ENSG00000000003",
"Symbol": "TSPAN6",
"TotalBases": 753,
"CoveredBases": 740,
"BasesCovered5x": 753,
"BasesCovered10x": 740,
"BasesCovered20x": 722,
"BasesCovered30x": 676
},
"ENSG00000000005": {
"ENSG": "ENSG00000000005",
"Symbol": "TNMD",
"TotalBases": 954,
"CoveredBases": 954,
"BasesCovered5x": 954,
"BasesCovered10x": 954,
"BasesCovered20x": 954,
"BasesCovered30x": 954
},
...